Re: [PATCH 1/5] Convert resource to u64 from unsigned long

[Andrew Morton <akpm@osdl.org>]

Dave <dave.jiang@gmail.com> wrote:
>
> ere's my first attempt of trying to convert the struct resource
> start/end to u64 per blessed by Linus =) This is to support >32bit
> physical address on 32bit archs such as some of the newer ARMv6 and
> XSC3 based platforms and perhaps IA32 PAE.  I left the PCI stuff alone
> functionally. Supporting 64bit PCI BAR on 32bit archs is for another
> day. I fixed most of the core stuff I can think of, fixed ARM and i386
> hopefully and a few of the device drivers as examples. I have tested
> on an IQ31244 XScale IOP (ARM) platform and a dual-xeon platform for
> i386. Matt Porter has graciously sent me PPC fixes that he tested on.

OK, well Greg KH will be the main target of this work..

Can you do something a bit more friendly than application/octet-stream
encoding, btw?

+#if BITS_PER_LONG == 64
+#define U64FMT "016lx"
+#else
+#define U64FMT "016Lx"
+#endif

We've avoided doing this.  We prefer to do

	printk("%llx", (unsigned long long)foo);

which is tidier, although a little more runtime-costly.

Your approach assumes that all 64-bit architectures implement u64 as
unsigned long (as opposed to unsigned long long, which I guess is legal?) I
don't know if that's a problem or not.

Also, the patches introduce tons of ifdefs such as:

+#if BITS_PER_LONG == 64			
	return (void __iomem *)pci_resource_start(pdev, PCI_ROM_RESOURCE);
+#else
+	return (void __iomem *)(u32)pci_resource_start(pdev, PCI_ROM_RESOURCE);
+#endif

We really should find a way of avoiding this.  Even if it is

#if BITS_PER_LONG == 64
#define resource_to_ptr(r) ((void *)(r))
#else
#define resource_to_ptr(r) ((void *)((u32)r))
#endif

in a header file somewhere.  Open-coding the decision all over the place is
unsightly.